In Yutong441/TBdev: Infer a Trophoblast Developmental Trajectory

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  warning = FALSE
)

Load data

library (TBdev)

We have processed 6 scRNAseq datasets and stored them under the data_dir as defined below. Each '.Robj' file contains a Seurat object for that dataset. Each dataset has already been normalised.

root <- '.' #in this case, it is the current directory
data_dir <- paste (root, 'data', sep='/')
save_dir <- paste (root, 'result', sep='/')

save_robj <- c('Xiang_R.Robj', 'Liu_R.Robj', 'Blakeley_R.Robj',
               'Petropoulos_R.Robj', 'Yan_R.Robj', 'Zhou_R.Robj')
# load all datasets as a single list of Seurat object
paper_ID <- gsub ('_R.Robj$', '', save_robj)
all_datasets <- load_all_data (save_robj, data_dir, paper_ID = paper_ID)
# This function will also create a metadata column 'paper' for the `paper_ID`

Merge data

The simplest way to integrate the datasets is to merge all expression matrix together, as implemented in merge_seurat function.

merged <- merge_seurat (all_datasets, 
                        assays = c('RNA')) #which assay to merge
merged <- run_dim_red (merged, 
                       find_var_features = T, #call `Seurat::VariableFeatures`
                       normalize=F, #no need to normalise the dataset again
                       var_scale=T, #scale the variable features for PCA
                       run_diff_map=F
)

# The cell type labels have been stored under the column 'Type'
# This step is extremely important for the correct order of appearance of cell
# labels in the figure legend and for assigning colors
merged <- clean_metadata (merged, cell_type_col='Type', date_col='date')
merged$cluster <- paste ('C', merged$seurat_clusters, sep='')

Inspect the merging process

AP <- list (pointsize=3, legend_point_size=3, fontsize=11, point_fontsize=4,
            font_fam= 'sans')
plot_dim_red (merged, group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='pca', AP=AP, further_repel=F)

In this case, the datasets seem to be merged well, as the two pre-implantation dataset (Petropoulos and Yan) are merged at a similar position. Likewise, Xiang and Zhou, two datasets on pre- and post- implantation trophoblasts, are merged together. The Liu dataset does not merge with others, probably because it contains first and second trimester placental samples.

plot_dim_red (merged, group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='pca', dims=1:3, AP=AP, further_repel=F, 
              all_theta=120, #degree of azimuthal rotation
              vert_just=0.9) #low the axis labels a bit

However, batch effects seem to be present from UMAP

plot_dim_red (merged, group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='umap', AP=AP, further_repel=F)

# reduce memory load
rm (merged)

Batch correction

Even though simple merging is successful in this dataset, let us investigate whether batch correction implementd in Seurat improves the integration process. Because Blakeley dataset contains fewer than 30 cells, which is lower than Seurat requirement, the dataset is removed here.

integrated <- batch_correct (all_datasets[1:6!=3], num_anchor_genes=1000)

rm (all_datasets)

Clean up the metadata

integrated <- clean_metadata (integrated, cell_type_col='Type', date_col='date')
integrated$cluster <- paste ('C', integrated$seurat_clusters, sep='')
integrated <- run_dim_red (integrated, assay='integrated', find_var_features =
                           T, normalize=F, var_scale=T, run_diff_map=F)

Note that the batch corrected data are in the 'integrated' assay

plot_dim_red (integrated,  
              group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='pca', AP=AP, further_repel=F, 
              move_y= 1, move_x=1) #shift the arrow axis to avoid view obstrucion

The batch effect seems to have been reduced significantly in UMAP. The UMAP matches well with lineage progression.

plot_dim_red (integrated, group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='umap', AP=AP, further_repel=F)

However, two key problems remain: the stromal cells (STR) overlap with syncytiotrophoblast (STB) and Oocyte and 4C overlap with cytotrophoblast (CTB). This is not because they are separated along the third dimension:

integrated <- Seurat::RunUMAP (integrated, assay='integrated', dims=1:30, n.components=3)
plot_dim_red (integrated, group.by = c('Type', 'date', 'paper', 'cluster'), 
              DR='umap', dims=1:3, AP=AP, further_repel=F, all_theta=20)